Skip to content

IBMFL 1.0.6
Compare
Choose a tag to compare
@chalianwar chalianwar released this 01 Aug 22:14
· 0 commits to ce-1.0.6 since this release
v1.0.6
757342d
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Learn about vigilant mode.

Release 1.0.6

IBMFL 1.0.6 includes several new functionality added, with improvements and bug fixes.

Changes

  • Support for Python 3.8.
  • Enhanced quorum support for the PFNM fusion algorithm.
  • Enhancement for FedAvgPlus fusion algorithm (previously named Fed+).
  • Refinement for library dependency.
  • Refactor of python notebook dashboard for the experimental manager.
  • Experiment manager dashboard now supports custom datasets.
  • No breaking changes.

New Functionality

  • Multi-cloud cluster support for OpenShift Orchestrator, see new openshift_fl folder.
  • New Shuffle fusion algorithm
  • Two variations of the Fed+ fusion algorithm, including Coordinate-median+ and Geometric-median+ (see new examples)
  • New Fed+ examples for PyTorch models and CIFAR10 dataset
  • Support to load and split custom datasets in csv format via generate_data.py -d <custom_dataset_name>

Bug fixes

  • Fix fairness datasets loading issues on aggregator and party sides
  • Correct Fed+ hyperparameter rho initialization
  • Fix default data path for id3_dt on Adult dataset example
  • Other minor fixes
Assets 2
Loading